Mike Conley (:mconley) (:⚙️) Reporter
	Description • 8 years ago

IME sends sync messages up to the parent process when focus shifts (see bug 1251063). This slows us from painting content when we, for example, first create a browser window or tab, since we'll be blocked on those sync messages for IME.

What I propose is that if TabChild gets RecvActivate called on it, it defers the activation until after it has first painted.

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 1 • 8 years ago

Actually, we probably want to wait until the composite has occurred.

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 2 • 8 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=24e37a48df39

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 3 • 8 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5a7e0b3a995c

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 4 • 8 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=d00808d1cc8a

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 5 • 8 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=22e8a0706cfa

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 6 • 8 years ago

So after talking with mstange, I think this might be a bit too ham-fisted. I wanted to defer activation until after paint/composite because of the sync IME messages that come up when the focus is switched to the browser in the new window in tpaint. Those sync messages take a lot of time.

What mstange pointed out to me was that if we defer activation, then any messages that are sent down to the content between first receiving the activation and actually _doing_ the activation after the paint, are not going to be interpreted correctly; we're breaking the assumption from the parent that if the content was sent Activate, that it's actually activated.

I think what we might want to do instead is figure out if we can somehow defer the IME sync messages coming up from the child instead of the activation message being reacted to by the child.

	Mike Conley (:mconley) (:⚙️) Reporter
	Comment 7 • 8 years ago
bugnotes

http://people.mozilla.org/~mconley2/bugnotes/bug-1258465.html